From: Sigbjørn Skjæret <redacted>
Date: Sun, 2 Aug 2026 07:03:05 +0000 (+0200)
Subject: model : load MiMo V2 MTP tensors only if used (#26412)
X-Git-Tag: upstream/0.0.10438~213
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=75587a05b39e5d5866a2e120178c75c12dfb75cb;p=pkg%2Fggml%2Fsources%2Fllama.cpp

model : load MiMo V2 MTP tensors only if used (#26412)
---

diff --git a/src/models/mimo2.cpp b/src/models/mimo2.cpp
index 4080a934c..d50e186cc 100644
--- a/src/models/mimo2.cpp
+++ b/src/models/mimo2.cpp
@@ -30,7 +30,11 @@ void llama_model_mimo2::load_arch_tensors(llama_model_loader & ml) {
 
     const std::string mtp_probe = "blk." + std::to_string(n_layer) + ".nextn.eh_proj.weight";
     const bool trunk_only = (hparams.n_layer_nextn > 0) && (ml.get_weight(mtp_probe.c_str()) == nullptr);
-    const int  mtp_flags  = trunk_only ? TENSOR_NOT_REQUIRED : 0;
+    int mtp_flags         = trunk_only ? TENSOR_NOT_REQUIRED : 0;
+
+    if (!ml.load_mtp) {
+        mtp_flags |= TENSOR_SKIP;
+    }
 
     tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);